This repository consists of training scripts modified from torchvision's vision repository to carry out a novel framework of Ensemble Structural Knowledge Distillation to train a Faster-RCNN model with a ResNet-18 backbone using two Faster-RCNN models, one with a ResNet-50 backbone and the other with a ResNet-101 backbone.
Post-Training Quantization was carried out to condense the model further to make it edge usable.
This research project was undertaken for my university course CSE465: Neural Networks and Pattern Recognition